A Dialogue between the Hypoxia-Inducible Factor and the Tumor Microenvironment

Through evolution, Mother Nature provided oxygen as a major resource for life. A cancer cell in close vicinity to a blood vessel is able to use the plentiful supply of oxygen carried by the blood to obtain energy through oxidative phosphorylation for rapid growth. Nonetheless, cancer cell oxygenation is limited by three convergent mechanisms: (1) As a tumor mass expands, cancer cells develop distant from blood vessels. The characteristic diffusion distance of oxygen results from the conjunction of the passive physical diffusion and active cellular oxygen consumption. Since this diffusion distance within tumor tissue is approximately 70 μm cancer cells proliferating at the periphery of blood vessels dispose of only a low oxygen availability [1, 2]. (2) Moreover, blood vessels in solid tumors are known to be highly disorganized and chaotic. This loss of coherent structural organization diminishes the oxygen profile of the tumor. To characterize the vascular network in tumors an original and very informative mathematical tool: the fractal character of the vascular system has been described [3, 4]. Histological studies have shown that the fractal dimension of blood vessels correlates with the nature of the tumor tissue. This dimension is higher in tumors than in non-malignant tissues, which reveals that the vessels present more irregularity and less homogeneity. Consequently perfusion and irrigation in tumors is often not optimal. (3) Finally the hematological status of cancer patients is frequently altered, either by the disease itself or by chemotherapy related toxicity. Hence these parameters converge to give inadequate irrigation of the tumor and thus low nutrient and oxygen availability. For example the median partial pressure of oxygen (pO₂) in the normal cervix is 42 mmHg compared to only 10 mmHg in squamous carcinomas [1, 2].

A tumor can be considered as a mosaic of blood vessels from which oxygen will diffuse into the tumor matrix, creating a myriad of gradients of oxygen concentrations within a hypoxic range. Clinical studies associate the hypoxic status of a tumor with bad prognosis and resistance to chemo- and radio-treatment [5]. Chemo-resistance can be explained partly by an inefficient diffusion of drugs to poorly irrigated areas due to an increase in internal fluid pressure. Furthermore, hypoxic radio-resistance is explained by the fact that radiotherapy uses oxygen to generate cytotoxic free radicals but also by active anti-apoptotic cellular mechanisms induced by hypoxia [6].

Hypoxia is a condition encountered in both physiological (embryonic development) and pathophysiological (ischemia diseases, diabetes, atherosclerosis, Alzheimer’s disease, chronic obstructive pulmonary disease, inflammatory disorders, pre-eclampsia, psoriasis and cancer) situations. Cells exposed to a hypoxic stress must rapidly adapt, otherwise an imbalance in their energy supply/consumption ratio ensues. Hypoxia activates a global signaling network centered on the key element, the hypoxia-inducible factor (HIF) [5, 7, 8]. Thus low oxygen tension is the prototypic regulator of HIF. This factor transcriptionally activates a panel of microenvironmental adaptation genes that contribute to rapid cell survival [5, 9, 10]. In addition to survival, the HIF signaling pathway confers on cancer cells an arsenal that allows them to become more aggressive. So, how is the transcriptional machinery activated in response to variations in the pO₂? HIF is a heterodimer composed of an O₂ regulated alpha subunit and a constitutively expressed beta subunit (Fig. 1). Several HIF alpha and beta isoforms have been described in mammals, where HIF-1α, HIF-2α and HIF-1β are the best characterized to date.

The basic helix-loop-helix (bHLH) domain in the N-terminal part of HIFα confers specific DNA recognition. DNA regions called hypoxia-response elements (HRE) are sequences characteristic of HIF downstream target genes. In hypoxia the heterodimer transactivates target HRE containing genes (1 to 2% of the genome is probably modulated by hypoxia). HIFα functional recruitment of RNA polymerase is driven by its transcriptional activation domains (TAD) where HIF-1α and HIF-2α share the distinctive feature of bearing two TADs: N-terminal (N-TAD) and C-terminal (C-TAD) (Fig. 1). The half-life of the HIFα protein in the presence of 21% oxygen is less than 5 min, and increases to 60 min as the O₂ concentration decreases to 1%. On reoxygenation, the HIFα protein is degraded after only a few minutes. As the O₂ concentration does not impact on the HIFα mRNA level, HIF is regulated exclusively posttranslationally by the pO₂.

The mechanisms by which cells sense and respond, through HIF, to environmental oxygen was revealed in the early 2000s. Dioxygenases catalyze protein hydroxylation, a reaction requiring ambient O₂, that leads to the formation of an −OH group on HIFα proteins. Oxygen-sensor dioxygenases target the HIFα subunit and are thus termed HIF-hydroxylases. Two types of HIF-hydroxylases have been characterized to date: (1) HIF-prolyl hydroxylases; though preferably referred to as prolyl hydroxylase domain (PHD) proteins since substrates other than HIF have been and are probably still to be identified, and (2) HIF-asparaginyl hydroxylase also called factor inhibiting HIF-1 (FIH) [11].

Three major PHD isoforms have been described and catalyze the hydroxylation of two prolyl residues of the HIFα subunit, an action that strictly depends of the pO₂ [12‐14]. These two residues are located in the oxygen-dependent degradation (ODD) domain of HIFα and when hydroxylated this domain shows a strong affinity for the von Hippel-Lindau (VHL) protein, a component of a E3 ubiquitin ligase complex. The consequence is poly-ubiquitination and targeting of HIFα for degradation by the proteasome (Fig. 2). Thus an increase in HIFα protein levels in hypoxia is due to the inhibition of this oxygen-dependent degradation process.

The second oxygen sensor FIH catalyzes asparaginyl hydroxylation in the C-terminal TAD of HIFα when in the presence of oxygen [11] (Fig. 2). This results in a change in the local hydrophilic/hydrophobic balance of the protein and impairs the interaction between the hydroxylated C-TAD and one of its essential coactivators: p300/CREB binding protein (CBP) [15, 16]. This abolishes the C-TAD activity, however, inhibition is progressively alleviated as the pO₂ decreases towards anoxia.

Thus, HIF hydroxylation is a double locking system. In normoxia HIF is degraded by highly active PHDs but if a pool of HIF escapes this system it will nonetheless be inactivated by FIH. When the oxygen tension decreases, HIF is stabilized and becomes fully active via its two TADs (Fig. 2).

Other forms of posttranslational modification of HIFα including phosphorylation, SUMOylation, S-nitrosylation and acetylation (the mechanism of the latter remains controversial, reviewed in [17]), have been reported to influence stability and activity [18]. Such modifications may also contribute substantially to the hypoxic response of cells depending on the environmental conditions.

Selective modulation of the spectrum of genes induced by HIF can occur in two ways. First, HIF-1α and HIF-2α can differentially transactivate a series of genes. For example adrenomedullin is a HIF-2 only gene in mouse stem cells, carbonic anhydrase IX (ca9) is a HIF-1 targeted gene in human HeLa cells, whereas phd3 is regulated by both HIF-1 and HIF-2 in human cells [19, 20]. Relative HIF-1 and HIF-2 activities differ depending on the cell type. Second, through the bifunctional TAD activity of HIF-1α [20] (Fig. 3a). FIH specifically hydroxylates the C-TAD, and does not touch the N-TAD. So experimentally, FIH inhibits only a subset of HIF-1-target genes. Knowing this we put forward a model that divides the HIF-1 spectrum of genes into two categories: one that is C-TAD sensitive and the other N-TAD only sensitive. FIH would inhibit the C-TAD spectrum and as a result pilot the shift between these two categories of HIF-1 target genes. This double TAD regulation model can be easily transposed to HIF-2α, since this isoform is regulated in a highly homologous mode.

To fully understand the impact of this model, these data must be considered in the context of an oxygen gradient within a tissue. How do oxygen-sensors drive HIF activity in the physiological context? Studies into the enzyme activity showed that PHD and FIH have very different requirements in terms of the oxygen concentration. In vitro, the Km of PHD for O₂ is three times greater than that of FIH [21]. This corresponds respectively to oxygen concentrations of approximately 24% and 8% (compared to 2% for collagen prolyl hydroxylases which are consequently insensitive to hypoxia). This means that PHD should be inactivated in milder (theoretically 3 times more oxygen) hypoxia compared to FIH. However, experimentally, in cellulo, FIH activity has been reported at 0.2% oxygen whilst HIF-1α is maximally stabilized by complete inactivation of PHD [22]. Thereby there should be areas of mild hypoxia in the tumor where HIFα is stabilized by the inactivation of PHD but still partially inhibited by FIH-dependent hydroxylation of its C-TAD (Fig. 3b). In such areas the N-TAD repertoire of HIF-dependent genes should be induced. Further, under severe hypoxia, FIH should also be inactivated, thus releasing the C-TAD sensitive repertoire. Thus this model provides a link between the induction of a certain gene at a certain localization and the oxygen-dependent activity of the oxygen sensor FIH (Fig. 3b). Moreover it confers a new and more subtle role to FIH which is no longer considered as a pure inhibitor of HIF but rather as a discriminator between two categories of HIF-dependent genes: N- and C-TAD targeted genes.

In summary, HIF gene induction is like a ‘four-handed concerto’ directed by two related but distinct transcription factors (HIF-1 and -2), each able to independently target genes with their two transactivation domains (N- and C-TAD). Yet this model still needs further investigation to fully appreciate the variation in the repertoire of genes when induced in a more physiological situation.

A number of HIF-induced genes directly influence environmental oxygenation (Fig. 4a). One of the best known is vascular endothelial growth factor (vegf-a) which favors the growth of new blood vessels into hypoxic zones [23]. Endothelial tip cells are capable of guiding the growth of blood capillaries towards hypoxic areas rich in VEGF-A. These endothelial tip cells emit long filopodia that are very rich in VEGF-R2 [24]. Hence the chemotactic gradient of VEGF-A is a guide for neo-vascularization to favor the sprouting of newly generated vessels toward poorly oxygenated areas. This phenomenon of angiogenesis, which is highly HIF-dependent [25], then substantially remodels the oxygen profile of the ischemic tissue or hypoxic tumor (Fig. 4b).

Another HIF-dependent molecule that is of great importance to blood vessel network remodeling is angiopoietin-2 (Ang-2) [26]. This protein antagonizes Ang-1 binding to its endothelial cell receptor Tie-2 and thereby prevents blood capillary maturation. The stabilization of the vascular network is dependent on activation of the Notch pathway, which results in quiescence of endothelial cells [27]. In this state, blood vessels are not sensitive to VEGF-A. So Ang-2, secreted from hypoxic regions of the tumor, by antagonizing its homologue Ang-1 destabilizes the organization of capillaries. This destabilization prior to blood sprouting is an essential step in making endothelial cells sensitive to VEGF-A and in initiating angiogenesis.

In addition to modulating extracellular oxygen homeostasis, HIF can impact on cellular respiration. It has been known for a long time that under low oxygen conditions cells shift from metabolism of glucose through oxygen-consuming oxidative phosphorylation (OXPHOS) in mitochondria to non-oxygen dependent glucose metabolism to lactic acid in the cytoplasm. This switch has now been revealed to implicate, at least in part, [28] HIF-induced glucose capture, glycolytic flux and inhibition of OXPHOS (Fig. 4b). Thus, HIF has a dual action on oxygen homeostasis. First, it restores, in the long-term, the oxygen supply through neo-vascularization and it promotes rapid reduction of O₂ consumption by reducing cell respiration.

Even if oxygen is the main and prototypic driving force of the HIF pathway, other potential HIF regulators have been identified. The glucose concentration in tumors follows a gradient that is very similar to that of the pO₂ gradient described above. The topologies of these gradients are intimately linked to each other. Glucose, like oxygen, is delivered by the blood circulation and diffuses into tissues for cell capture.

The idea that microenvironmental nutrients could influence HIFα originated from the fact that the enzymatic reaction catalyzed by oxygen-sensors is dependent on the metabolite 2-oxoglutarate (2-OG). In vitro studies show that 2-OG is a co-substrate for HIF prolyl and asparaginyl hydroxylation by respectively, PHD [12] and FIH [11]. Thus, the PHD and FIH activities could be impaired by reduced levels of 2-OG. Since 2-OG is a product of the tricarboxylic acid (TCA) cycle (also Krebs cycle) and thus of glucose metabolism glucose availability could influence activity. The normal level of 2-OG ranges from 50 et 230 μM [29] while the in vitro Km for the PHDs has been estimated to be about 60 μM [30]. This level of affinity would classify the PHDs as nutritional sensors. Thus, through 2-OG the PHDs are sensitive to TCA cycle activity and the balance between aerobic and anaerobic metabolism.

A decrease in the amount of glucose provided to hypoxic cells was shown to result in a decrease in the level of HIF-1α [31] or to have no effect [32]. The explanation for this difference may lie in either the difference in cell lines or the level of hypoxia (0.1 or 1.0% oxygen) used. Thus, the sensitivity to glucose in vivo remains to be further clarified.

Another type of nutritional stress, amino-acid depletion, is able to decrease HIF protein expression by activating the adenosine monophosphate kinase (AMPK) and mammalian target of rapamycin (mTOR) pathway (Fig. 5a). In a situation of an energy imbalance and decrease in intracellular ATP, the tumor suppressor serine/threonine kinase LKB1 phosphorylates its downstream effector AMPK that has the potential to target the tuberous sclerosis complex (TSC1/TSC2) [33]. Once activated, the TSC1/TSC2 complex inhibits mTOR resulting in a decrease in protein synthesis. Even if it is generally accepted that a decrease in mTOR activity has a negative effect on HIFα expression, the mechanisms are controversial. Indeed, HIF protein synthesis has been shown to be upregulated by mTOR in various cell types [34, 35]. Genetic evidence reinforces this link at the level of translation between mTOR and HIF, by the presence of a 5′-terminal oligopyrimidine tract (5′-TOP) sequence in the 5′-untranslated region (5′-UTR) of HIF-1α. 5′-TOP sequences can be driven by the S6 ribosomal protein that is itself a downstream target of mTOR via p70S6Kinase. Yet it has been demonstrated that mTOR does not change HIF protein synthesis but modifies stabilization of HIFα in prostate PC3 cells [36]. This result is consistent with the fact that the HIF 5′UTR has been demonstrated to bear an internal ribosome entry site theoretically allowing HIFα to be translated even in the absence of mTOR driven CAP dependent translation [37]. Moreover, evidence indicates that HIF protein synthesis can be controlled by stimulation of an Akt-dependent but mTOR-independent pathway in PC3 cells [38]. To complete the picture, a feedback loop allows HIF to downregulate mTOR via the hypoxia inducible REDD-1 protein by activating the TSC1/TSC2 signaling integrator complex [39]. In conclusion, a close link exists between mTOR and HIF, bringing together two fundamental microenvironmental constraints: nutrient and oxygen, which are central to generation of cellular energy.

Nutritional deprivation induce a cellular catabolic response termed autophagy that can rescue cancer cells from cell death. Autophagy is a phenomenon where proteins, organelles and cytoplasm of a cell are engulfed into vacuoles and degraded into constituents for recycling for maintenance of metabolism and thus cell viability. Thereby, in tumor cells where apoptosis is defective autophagy promotes cell survival. However, it is paradoxical that autophagy is associated with increased tumorigenesis by a mechanism that remains to be defined. It has been suggested that autophagy protects cells from tumor necrosis and inflammation, and diminishes DNA damage in the tumor cell response to metabolic stress [40].

While HIF is a major actor in the cell survival response to hypoxia, HIF has also been associated with cell death. Indeed several studies have pointed to the implication of the HIF-induced putative BH3 only pro-apoptotic genes Bcl2/adenovirus E1B19kD protein interacting protein 3 (bnip3) and bnip3 like (bnip3L) in hypoxia-mediated cell death [41, 42]. We and others failed to reproduce the pro-apoptotic or necrotic cell death features of ectopically expressed BNIP3 or BNIP3L in various cell types including MEFs, MCF7, PC3 or LS174 cells. Therefore, we seriously questioned the cell death function of BNIP3/BNIP3L in a hypoxic environment and rather postulated on a positive role in activating the autophagic cell survival process [43]. Using a large panel of normal cells and tumor cell lines, we demonstrated that hypoxia-induced BNIP3 and BNIP3L (1% O₂) were required to initiate autophagy. Whereas siRNA-mediated ablation of either BNIP3 or BNIP3L had little effect, combined silencing of the two HIF targets suppressed hypoxia-mediated autophagy. BNIP3 and BNIP3L are therefore major players of the HIF-mediated survival response in tumors and ischemic tissues. We propose a model in which the hypoxia-induced BNIP3/BNIP3L proteins are essential to disrupt the Bcl2/Beclin1 complex ([44], Mazure et al. in preparation). It is interesting to note that autophagy is triggered in conditions of full availability of nutrients (glucose, amino acids) and growth factors; the only limiting factor is oxygen. We propose that the pO₂ is, as for angiogenesis, the primary signal capable of preparing the tumor cell to endure prolonged nutrient starvation by inducing autophagy.

From a metabolic view point, the cellular energetic balance of hypoxic tumor cells is compromised by the shift away from OXPHOS, as mentioned above, since the alternative i.e. conversion of glucose to lactate produces less energy. To make up for lower ATP production tumor cells increase the capture of glucose and the flux of glycolysis through HIF-mediated induction of genes of glucose transporters such as glut-1and glut-3, and glycolytic enzymes like pgk1 (Fig. 5b). In addition, HIF redirects the fate of pyruvate as a result of two concordant actions: (1) by upregulating the gene lactate dehydrogenase-a (ldh-a), the protein product of which favors the conversion of pyruvate to lactate [45] (2) by repressing OXPHOS through induction of the expression of pyruvate dehydrogenase kinase-1 (PDK-1), an inhibitor of pyruvate dehydrogenase (PDH), the enzyme that drives pyruvate into the TCA cycle for mitochondrial respiration [46, 47]. Finally HIF impacts on OXPHOS by modulating cytochrome c oxidase (COX) isoform expression [48]. In fact, oxygen is the final acceptor in the mitochondrial electron transport chain and COX, or more precisely COX-4 of complex IV, catalyzes the reaction with production of H₂O [49]. When this reaction is not sufficiently active oxygen is the final electron at the level of complex I or III that produces reactive oxygen species (ROS). The expression of COX4-1 and COX4-2, two isoforms of complex IV, is differentially regulated by HIF through the up-regulation of the mitochondrial protease LON since LON in turn degrades COX4-1 [48]. Thus, in hypoxia COX4-2 substitutes for COX4-1, an isoform that is suggested to be more efficient in oxygen utilization. In this way cells exploit more efficiently low oxygen levels and avoid production of harmful ROS under hypoxic conditions.

Another important feature of the tumor microenvironment is acidosis, which is a consequence of the boost in glycolysis leading to production of lactic acid and of defective vascular evacuation of metabolic lactic acid and CO₂ [50]. As for oxygen and nutrients, a gradient of acidosis is established from blood vessels to the periphery of a tumor. The decrease in oxygen and pH was examined along a length of 300 μm from blood vessels in a breast cancer xenograft mouse model and a drop in pO₂ from 12 to 0 mmHg and in extracellular pH (pHe) from 7.4 to 6.7 was reported [51]. The question as to whether pH can influence the HIF pathway was first examined through an indirect effect on the VHL protein, a component of the HIFα ubiquitin ligase complex [52] (Fig. 6a). Acidosis was shown to induce relocalization of the VHL protein into nucleoli. However, it is still not clear if a drop in pH on its own, is sufficient to allow HIF to bypass polyubiquitylation and thus escape proteasomal degradation. Results from our laboratory show that a low pH is not necessarily associated with HIFα stabilization (unpublished results), which was corroborated in a recent publication [53].

Low pH has also been shown to modulate two HIF-induced genes: vegf and bnip3. Both pH and hypoxia can independently induce vegf in human glioma cells [54] and the extracellular-regulated kinase (ERK) pathway is an independent intermediary candidate up-regulating the level of vegf in a pH-sensitive manner [55]. Although low pH does not increase BNIP3 protein expression it increases its activity in regulating cell death in cardiomyocytes [56]. It is possible to hypothesize that HIF-1 induces BNIP3 in moderate hypoxic regions whereas the pH gradient pilots BNIP3-directed cell death only in highly acidic/hypoxic areas [43]. Thus, low pH might be a major contributor in sensitizing cells to necrotic cell death. Nevertheless the link between HIF and cell death is a highly debated subject and the precise conditions of hypoxia-induced cell death in tumors still needs to be clarified [57].

Thus, hypoxia and acidosis cohabit in tumors but both undermine cell survival. In response, HIF triggers an adaptation strategy that leads to induction of specific genes dedicated to pH homeostasis (Fig. 6b). The gene products include the Na⁺/H⁺ exchanger (NHE) that extrudes protons from the cytoplasm at the expense of the Na⁺ gradient and the monocarboxylate transporter (MCT) that evacuates lactic acid. The expression of the MCT4 [58], NHE1 [59] and NHE6 [47] isoforms is induced in hypoxia and the exchange activity of NHE1 is increased [58, 60]. MCT can transport lactate in both the intracellular and extracellular direction. Thus, lactate may be considered as both an acidic waste product and a source of energy, as has been shown for muscles and hypothesized for neurons [61]. In a tumor it may be envisaged that a cancer cell excretes lactate to be taken-up by neighboring fibroblasts for subsequent use by the TCA cycle. In fact, it has been shown that fibroblasts in contrast to cancer cells express a high level of PDH and a low level of PDK1 [62]. Thereby fibroblasts through an oxidative utilization of lactate may promote extracellular lactate clearance. The intracellular transport of lactate by endothelial cells that express active MCT1 [63] in tumors may also modify the microenvironment through promoting angiogenesis [64].

In addition, the membrane bound carbonic anhydrases (CA) CA IX, one of the most HIF-sensitive genes, and CA XII convert environmental CO₂ into bicarbonate which may alkalinize the intracellular pH (pHi), possibly through capture by Na⁺-dependent and -independent Cl^-/HCO³⁻ exchangers [65].

The extracellular matrix (ECM) indirectly impacts on all these environmental characteristics. Oxygen and nutrient diffusion are dependent on the stroma density, vasculature and cellular three-dimensional organization. Beside these physical characteristics, the chemical composition of the stroma can also be a determining feature (Fig. 7a). For instance, growth factors can modulate the HIFα protein level. Insulin like growth factor 1, epithelial growth factor and epithelial growth factor 2 (HER2/Neu) have been shown to induce HIF-1α expression in respectively, colon carcinoma cells (HCT116) [66], prostate cancer cells (DU145, PC-3, PPC-1, and TSU) [67] and breast cancer cells (MCF-7) [35]. These three factors increase the HIFα protein level via the phosphatidylinositol-3-kinase (PI3K)/Akt pathway targeting mTOR. However, the ERK pathway is also a candidate for HIF modulation by growth factors since HIF activity is stimulated through ERK-dependent signaling [68]. At least two mechanisms might explain HIFα up-regulation by ERK: the co-activator p300 that enhances HIF activity by binding to the C-TAD of HIF may be targeted by phosphorylation leading to a more favorable p300-HIF-TAD interaction [69], and/or HIF targeted phosphorylation may promote HIF-1α nuclear localization [70]. In addition, vasoactive hormones such as angiotensin II that are secreted by stromal cells can promote both HIF translation in a PI3K-dependent manner and HIF transcription through the action of the diacylglycerol-sensitive protein kinase C in vascular smooth muscle cells [71].

A series of hypoxia induced phenomena impact on the ECM. Indirectly, hypoxia-related acidosis can modulate the composition and architecture of the ECM, which in turn impacts on three-dimensional cellular organization and promotes metastasis (Fig. 7b). In a more direct fashion, hypoxia, through HIF is a modulator of lysyl oxidase (LOX) [9, 72], an enzyme that catalyzes collagen and elastin crosslinking [73]. Furthermore, stimulation of LOX by HIF is an essential intermediary of hypoxia-promoted metastasis [72, 74]. Surprisingly the canonical action of LOX on collagen and elastin might not be the key to this pathway but rather peroxide production, a side product of LOX activity, and focal adhesion kinase activity may be at the center of this pathway. However, the question as to whether LOX is associated positively or negatively with tumor development has hardly been discussed [72, 74‐76], yet this ECM modifier is strongly induced by hypoxia. The significance of LOX in cancer should be reconsidered taking into account the hypoxic status of a given tumor.

Another family of proteins that modifies the ECM and that is directly involved in metastasis is the matrix metalloproteases (MMP). These proteins participate in the destruction of the ECM, an action known to be essential for cellular migration. MMP-2 and membrane type 1-MMP1 expression [77, 78] has been shown to be responsive to HIF-1 and HIF-2, respectively. In the same context, metabolic changes and HIF action have been shown to play a role in adhesion, in particular in interaction with vascular endothelial cells through a selectin- and integrin-mediated pathway [79]. Evidence for HIF negative regulation of E-cadherin might also provide another clue to elucidate how cancer cells lose contact and proceed to epithelial to mesenchymal transition leading to metastasis [80‐82]. This process is particularly relevant in renal cell carcinoma where the loss of the ubiquitin ligase VHL protein function correlates with overexpression of HIF. Nonetheless the intermediary signal linking HIF and E-cadherin remains to be identified. Another group of HIF-dependent genes impacting on the remodeling of the ECM include, fibronectin, cathepsin D and urokinase plasminogen activator [5]. Yet another series of genes products also induced by HIF has been shown to promote cell invasion: the autocrine motility factor [83], vimentin, the receptor tyrosine kinase c-Met [84], the stromal derived factor-1, keratins 14, 18 and 19 and the cytokine receptor CXCR4 [85, 86].

Finally, the inflammatory status of the tumor stroma is a crucial element of the microenvironment and the HIF pathway can modulate the cellular inflammatory response. On the one hand, by promoting glycolysis, lactate production and thus extracellular acidification, HIF indirectly inhibits a subset of inflammatory actors: human cytotoxic T lymphocytes [87] (which are sensitive to lactate) and natural killers [88] (inactivated at low pH). On the other hand, HIF is a powerful driving force behind the action of tumor associated macrophages (TAM). As a result of over activation of the HIF pathway that increases glycolysis TAM are supplied with the energy necessary to migrate to tumor sites [89]. Consequently, TAM have a propensity to move toward areas distant from the circulation such as hypoxic and peri-necrotic zones [90]. In addition, they contribute to tumor aggressiveness through their ability to promote angiogenesis, extracellular matrix remodeling and cancer cell migration [91, 92].

This large body of data reveals the global role of HIF in tumor cell adaptation to microenvironmental stress. Commencing with a decrease in the pO₂ as the main regulator, this transcription factor initiates a cascade of events that modifies profoundly the microenvironment: tissue perfusion, metabolism, acidosis, matrix remodeling, proteolysis, and cellular migration. Taken together these data place HIF at the center of potential anti-environmental directed therapies.

Based on the fact that O₂ is the main regulator of the HIF pathway, and that clinically hypoxic tumors are associated with poor patient survival, interfering with oxygen homeostasis constitutes an interesting anti-cancer approach. However, hypoxia is both a cytotoxic stress and a stimulus in initiating adaptation. The balance between the good and the bad side of hypoxia is very fine and might be tumor type dependent.

An illustration of the difficulty in foreseeing and controlling the precise effects of oxygen can be found in the controversy touching hyperbaric oxygen therapy (HBO). This type of treatment consists in submitting patients to 100% oxygen at a pressure higher than one atmosphere. The idea is to decrease the hypoxic score of the tumor. Contradictory results have emerged both from clinical investigation and animal experimentation [94]. HBO therapy can be pro- or anti-proliferative, or even inactive. On the one hand, oxygenation decreases the HIF microenvironmental adaptation response, but on the other hand it brings precious fuel to cancer cells.

The goal of anti-angiogenic approaches is to disturb vessels and consequently the oxygen profile in the tumor. By inhibiting VEGF [23, 95] or the endothelial VEGF-receptor [96‐98] the formation of new vessels, a crucial process in the dynamics of tumor growth, is impaired. However, this approach paradoxically has limitations since anti-angiogenic agents will increase the hypoxic score of tumors, which is likely to increase cell virulence (Fig 8).

Another strategy consists in considering oxygen as a landmark of the HIF adaptive pathway where oxygen is a kind of environmental messenger dictating HIF function. An interesting strategy would be to specifically target hypoxic cells. In fact, their location, distant from blood vessels contributes physically to their resistance to classical chemotherapy (delivered by the circulation) and radiotherapy (that uses oxygen as an intermediary target to generate toxic free radicals). Hypoxic cells are thus particularly strategic targets. Compounds such as tirapazamine (TPZ) [99‐101] are pro-drugs that become activated by the reducing effect of a drop in oxygen tension. Therefore their toxicity is targeted to hypoxic areas in tumors. However, the question of diffusion of this kind of drug from blood vessels still needs to be addressed. Yet it is conceivable that their lower non-specific toxicity should allow high dosage. Future studies should evaluate the exact sensitivity and toxicity of this kind of treatment. Interestingly the activity of the HIF pathway has been shown to decrease TPZ activity probably due to the redistribution of intracellular oxygen following mitochondrial inhibition [47]. This recent data suggests that a combined anti-HIF and TPZ therapy should clearly improve treatment. Alternatively, the hypoxic characteristic of tumors can be exploited in therapy by using obligate anaerobic bacteria such as clostridia, which show lytic activity in tumor cells [102].

The hypoxic status of a tumor might also be considered as a key parameter in the choice of the anti-cancer arsenal of physicians. Susceptibility of a given drug could indeed be oxygen dependent [103]. For instance this seems to be the case for the proteasome inhibitor bortezomib [104] and for mTOR inhibitors [105]. Thus the hypoxic score might not only be predictive of the tumor aggressiveness but also guide the clinician in the therapeutic strategy to adopt.

Clinical evidence suggests that HIF-1 overexpression is associated with higher cancer mortality or treatment resistance and thus inhibition of HIF activity has been proposed [5]. A number of HIF modulators that target different steps in HIF function are currently under evaluation [5, 103, 106]. However, the benefit of such a therapeutic approach remains speculative since HIF is a pleiotropic factor, able to promote survival, proliferation, and cell death, thus modulating its action might have drastic consequences on general physiology. Thus, instead of targeting HIF itself, targeting a particular set of HIF downstream products may show potential. However, this strategy might suffer from the complexity and potential redundancy of the HIF pathway. Even if a precise adaptation gene is inactivated, the relay may be taken up by one of the myriad of HIF downstream targets. The expected cytotoxic effect could possibly be circumvented thanks to another microenvironmental adaptation phenomenon. Interrupting the dialogue between HIF and the microenvironment certainly appears to be complex and requires an in-depth understanding of the mechanisms involved in regulation.

Agents such as echinomycin impair HIF-1 association with its target DNA [107] while growth factor directed therapies such as traztuzumab (Herceptin), which inhibits HER2/Neu, indirectly decrease HIF-1α protein levels [35, 106]. Benzoquinone ansamycin drugs like geldanamycin, also decrease the HIF-1α protein level by favoring its proteasomal degradation [108]. Alternately, interest in developing HIF activating drugs in the treatment of ischemic disorders has led to research into agents such as dimethyloxaloylglycine which inhibit the 2-OG dependent dioxygenase activity resulting in stabilization and activation of HIFα [109].

As mentioned, selective inhibition of the appropriate HIF-downstream gene product may be preferable. VEGF and VEGF-R are such targets and although precursors in this approach they may not be the only HIF downstream targets of interest. The combination of the anti-angiogenic with anti-HIF metabolic adaptation agents could be envisaged by interfering with key glycolytic enzymes such as LDH-A. Another strategy would consist in impairing the HIF-induced pHi regulating system [43]. In this case, the natural propensity of cancer cells to switch to pyruvate metabolism to lactic acid and consequently to acidosis would impact on proliferation. As a proof of principle, genetically altered cell lines that are not able to regulate their pHi do not proliferate normally under acidic conditions and show a very poor tumorigenic potential in xenograft experiments [110]. From a therapeutic point of view, the expected cytotoxicity would result from necrotic cell death sensitized by intracellular acidosis.

A third approach, which is a compromise between the previous ones, would be to exploit the HIF pathway by destabilizing its global pro- and anti-survival balance in favor of tumor regression. This would consist in targeting a qualitative modulator of the HIF activity. In this context, FIH would be an interesting candidate. Indeed, we discussed above the potential role of FIH as a discriminator between two subtypes of genes and as a determinant of the gene localization in a tumor hypoxic gradient [20]. The FIH activity would direct the expression of a precise gene at a precise localization in the tumor, associated with a given oxygen tension. It is likely that genes associated with cell death and anti-proliferation are activated in severe hypoxic regions whereas survival genes are induced in milder hypoxic zones. Displacing anti-survival genes to oxygenated regions might be of great interest. By targeting FIH, our working model predicts such a disorganization of the whole gene pattern that we believe should impair the general microenvironmental adaptation process.